I recently wrote following question to the dt users ml, but didn’t get a reply so far. Hoping that maybe different folks are reading here, I’m asking for help again here. Any reply here or on the ml is highly appreciated.
Dear ML members!
I’m trying my first steps with the dual iso feature of the magic lantern firmware for my Canon EOS 5D Mark III. Running Ubuntu 18.04 with darktable 2.4.4 I compiled cr2hdr and use the lua script[1] to convert them. The result looks a little strange to me, having zebra like stripes in the middle tones. Find a picture + xmp + screenshot under following link: https://www.dropbox.com/sh/rlr5az3hq333pl7/AABJLZmuTEgWPEwnhIE_EVh9a?dl=0
Is this intended, what did I do wrong/miss? Any hints appreciated.
I haven’t used it in a long time. I don’t remember the particulars. However, one way to get started is to tell us what camera and ML settings were used. Usually, it is the user’s issue, not the software’s. That may be why the ML folks aren’t responding. Or you haven’t given them sufficient time.
Dual ISO means that two ISOs are being used in a alternate fashion, so the highlights and shadows suffer from half resolution but the overall image has higher dynamic range. Given that information, it is quite possible for there to be this striped pattern. However, cr2hdr should have processed that.
(In the case of the moon, I would not have considered using Dual ISO.)
As written above I’m using a Canon EOS 5D Mark III. The picture was taken with ISO 800 (see screenshot or file at the link I posted before). In ML I’m using following settings:
Recovery ISO: +3EV
Alternative frames only: ON
Custom file prefix: DUAL (unreliable!)
Not sure what additional information you ask for.
Well because I believe it’s not a bug but a handling error I did write to the mailing list and did not create an issue at the bug tracker… I’m happy to be told what I did wrong!
That’s what I believe as well, so what did I do wrong?
What is your other ISO, DR gained and mid-tone overlap?
Correction below: I reversed “base” and “recovery” ISO.
Chances are that 1. You may have selected the wrong ISO pair for the job. To me +3EV is nonsense, try picking a recovery ISO value (in ML) that is good (i.e., in terms of SNR) for your night shot and then choose a base ISO (in camera) that is good for live view. If you use the viewfinder, then it doesn’t matter which is which.
2. You may have a low DR gained and / or a low mid-tone overlap. Low mid-tone overlap is the likely problem to your woes. The greater the difference in values between the ISO lines, the higher the chance of artifacts.
Since the moon is such a small yet bright subject in the sky, it is likely that it would be suspect to both the half resolution and the artifacts.
I use RT, not DT, but I remember encountering a similar issue a long time ago when I tried using Dual ISO for a photograph of the moon. I remember solving it by changing the demosaicing algorithm from AMaZe to something else (but I don’t remember which algorithm I used).
I was using PPG, the default of darktable. Changing it to AMaZE seemed to not make much difference, changing it to VNG4 decrease the contrast of the grey stripes a little – but they’re still dominant visible, and the yellow and green ones obviously even more.
I did understand that it might be helpful to decrease the difference between the two ISO values (meaning to decrease the +3EV) and additionally to maybe choose a different pair of ISO values all together (if the scene/exposure time) permits. For the used Canon EOS 5d Mark III one reasonable ISO value seems to be ISO 3200 (maybe ISO 6400 using the zoom lens with extender).[1] I’m not totally sure (especially after reading You probably don’t know what ISO means – and that’s a problem) if my other ISO value should be lower or higher though.
I’ll anyway need to wait a couple of years to try different settings as recommended for the next lunar eclipse photos. But I’ll continue to experiment and come back if I have additional questions.
This is what I meant: you should use the actual ISO value instead of EV while paying attention to what happens to DR gained and mid-tone overlap values. +3EV doesn’t mean anything in of itself. If you played around with the two ISO values a bit more, you would realize that DR gained depends which specific ISOs you chose.
Again, let me elaborate on my previous post.
1. Choose an ISO that you would use if you didn’t use Dual ISO; i.e., to get your optimal SNR, or whichever way you measure acutance.
2. Select an ISO buddy that is less than that ISO.
Why is that? It doesn’t matter if we misunderstand what ISO is, etc., the fact is that a lower ISO normally has a high DR, which is the whole point of Dual ISO. Try choosing an ideal ISO and then pair it with a lower one and compare that with a higher pair. Chances are that DR gained would be higher while the overlap is the same.
Hope this helps. In the end, you have to be a tinkerer.
